Semantic Reinforced Attention Learning for Visual Place Recognition